Dynamic planning under fundamental uncertainty – AI for decision-making
About the project
Realistic problem environments often undergo rapid and unpredictable change. It’s crucial that machines are programmed to adapt under such conditions to avoid the possibility of reaching failure states at any time.
Using regular convergence methods that accumulate and interpret past data to predict average future values to achieve a fixed ideal solution is inadequate. Planning under uncertainty requires abstract properties that remain constant despite the unpredictability of future state pathways.
This project aims to develop a dynamic approach to building optimal strategies in uncertainty, where plans are not defined before any interactions take place, but instead, while interacting and learning within an unknown environment. Further, we will investigate how to adjoin ‘on-the-go' learning to state-of-the-art, logic-based and automata-theoretic approaches.
Part of the AI for Decision-Making (AI4DM) Initiative, a collaboration between the Office of National Intelligence and the Defence Science and Technology group, this project aims to enhance AI and machine learning expertise in areas of focus for Australian defence and national security. The work will improve the way AI agents of any kind make optimal decisions in autonomous ways to benefit the country’s safety.